The Welfare Effects of Encouraging Rural-Urban Migration
David Lagakos
Mushfiq Mobarak
Michael E. Waugh
UCSD and NBER
Yale University
NYU and NBER
May 19, 2017
0 / 51
Misallocation and Cross-Country TFP Differences
• Cross-country differences in income per capita accounted for largely by
TFP, i.e. “a measure of our ignorance” (e.g. Caselli, 2005)
• Misallocation as a theory of TFP: factors of production inefficiently
allocated across firms or sectors
• Hsieh & Klenow, 2009; Restuccia and Rogerson, 2009 + enormous
follow-up literature
1 / 51
Rural-Urban Income Gaps = Worker Misallocation? Large urban-rural living-standards gaps in cross section • Value added per worker (Gollin, Lagakos and Waugh, 2014) • Consumption measures (Young, 2014) • Wages (Herrendorf and Schoellman, 2015)
2 / 51
Rural-Urban Income Gaps = Worker Misallocation? Large urban-rural living-standards gaps in cross section • Value added per worker (Gollin, Lagakos and Waugh, 2014) • Consumption measures (Young, 2014) • Wages (Herrendorf and Schoellman, 2015)
Rural-urban migrants increase their consumption • Bryan, Chowdhury, Mobarak (2014): Induced seasonal migrants
experienced 30% increase in consumption. • In tracking data: e.g. Beegle, De Weerdt, Dercon (2011) in Tanzania;
LSMS Panel Surveys from Tanzania and Uganda
2 / 51
This Paper: Welfare Effects of Subsidizing Rural-Urban Migration Draw on unique evidence from Bangladesh • 156 million people, $900 per capita • Urban wages 80% higher than rural • Fall “lean season”: very low productivity in rural areas
3 / 51
This Paper: Welfare Effects of Subsidizing Rural-Urban Migration Draw on unique evidence from Bangladesh • 156 million people, $900 per capita • Urban wages 80% higher than rural • Fall “lean season”: very low productivity in rural areas
Migration experiment of Bryan, Chowdhury & Mobarak (2014) • 100 randomly selected villages in Rangpur region, in NW Bangladesh • Treatment group offered 600 Taka ($8.50) conditional on seasonal
migration; 200 Taka extra if checked in at destination • Migration rates 58%, compared to 36% in control group • Among those offered treatment and migrated, consumption 30% higher
3 / 51
60
Seasonal Migration Rates in Bryan, Chowdhury & Mobarak (2014)
20
Control
Difference
0
Migration Rate
40
Incentivized
2008
2009
2010
2011
2012
2013
4 / 51
What We Do • Build two-region model with migration, featuring:
1. Sorting by comparative advantage (Roy, 1951) 2. Migration disutility that depend on migration experience 3. Incomplete markets as in Bewley-Huggett-Aiyagari - Differential urban & rural risk (Harris & Todaro, 1971) - Credit constraints as in development literature
5 / 51
What We Do • Build two-region model with migration, featuring:
1. Sorting by comparative advantage (Roy, 1951) 2. Migration disutility that depend on migration experience 3. Incomplete markets as in Bewley-Huggett-Aiyagari - Differential urban & rural risk (Harris & Todaro, 1971) - Credit constraints as in development literature • Novelty: discipline model using experimental evidence
- Perform experiment within model - Match experimental moments using simulated method of moments
5 / 51
What We Do • Build two-region model with migration, featuring:
1. Sorting by comparative advantage (Roy, 1951) 2. Migration disutility that depend on migration experience 3. Incomplete markets as in Bewley-Huggett-Aiyagari - Differential urban & rural risk (Harris & Todaro, 1971) - Credit constraints as in development literature • Novelty: discipline model using experimental evidence
- Perform experiment within model - Match experimental moments using simulated method of moments • Use model to interpret experiment, compute welfare gains 5 / 51
Summary of Results • To match data, model requires:
- Few rural households with strong comparative advantage in city - Substantial non-monetary disutility of migration that’s “persistent” - Seasonal migration mostly by those with lowest income and assets
6 / 51
Summary of Results • To match data, model requires:
- Few rural households with strong comparative advantage in city - Substantial non-monetary disutility of migration that’s “persistent” - Seasonal migration mostly by those with lowest income and assets • Welfare gains mostly from targeting funds to poor, vulnerable households
- 0.4% on average; 1.5% for poorest - Better for poor than unconditional transfer program costing same
6 / 51
Summary of Results • To match data, model requires:
- Few rural households with strong comparative advantage in city - Substantial non-monetary disutility of migration that’s “persistent” - Seasonal migration mostly by those with lowest income and assets • Welfare gains mostly from targeting funds to poor, vulnerable households
- 0.4% on average; 1.5% for poorest - Better for poor than unconditional transfer program costing same • Higher welfare gains in principle when, counterfactually, there is:
- Lower disutility of migration / more ”misallocated” rural workers
6 / 51
Summary of Results • To match data, model requires:
- Few rural households with strong comparative advantage in city - Substantial non-monetary disutility of migration that’s “persistent” - Seasonal migration mostly by those with lowest income and assets • Welfare gains mostly from targeting funds to poor, vulnerable households
- 0.4% on average; 1.5% for poorest - Better for poor than unconditional transfer program costing same • Higher welfare gains in principle when, counterfactually, there is:
- Lower disutility of migration / more ”misallocated” rural workers • New “discrete choice” experiment: housing key component of disutility 6 / 51
Model
7 / 51
Households and Preferences Unit mass of households; preferences: ∞ X
β t u(ct )u¯xt
t=0
• u(ct ) = ct1−α /(1 − α) • u¯ ∈ R is disutility of migration • xt ∈ {0, 1}, takes on value 1 iff household is “inexperienced at migration”
and in the urban area
8 / 51
Experience at Migration • A household is either “experienced” or “inexperienced” at migration I
Could reflect e.g. relationship with an employer
• After each period in the urban area I
Inexperienced households remain so with probability λ, and become experienced with probability 1 − λ
I
Experienced households stay experienced
• After each period in the rural area, I
Experienced households stay experienced with probability π and loose their experience with probability 1 − π.
I
Inexperienced households stay inexperienced
9 / 51
Production Technologies and Seasons One homogenous good produced in both locations, competitive Production technology in rural area: Yr = Ari Nrφ where 0 < φ < 1; Ari is rural productivity indexed by season i, with Arg > Arl • Deterministic transition: If rural productivity is Arg , then the economy
transits to Arl next period. • Idea: mimic seasonal agricultural crop cycles
Production technology in urban area Yu = Au Nu
10 / 51
Household Location-Specific Productivity Each household has permanent productivity draw as in Roy (1951) • z = permanent productivity in urban area. z ∼ Pareto(θ). • all workers equally productive in rural area.
Each household experiences transitory shocks s that follow an AR(1) process: log st+1 = ρ log st + t+1
with
t+1 ∼ N (0, σs ).
Household-specific efficiency units of labor in each location: • s in the rural area. • zs γ in the urban area. • γ > 1 implies the shocks are more volatile in the urban area.
11 / 51
Location / Migration Options Households in the rural area can. . . 1. Work in the rural area. 2. Pay fixed cost mT , work in the urban area the next period, return to rural. This is seasonal migration. 3. Pay fixed cost mp > mT , move to urban area next period, stay indefinitely. This is permanent migration. Households in the urban area can. . . 1. Work in the urban area. 2. Pay fixed cost mp , work in rural area for the indefinite future.
12 / 51
Asset Choices Households can accumulate a non-state contingent asset, a Gross rate of return, R. Assets move with the household. For today. . . • Asset holdings can not be negative (no borrowing) • R is exogenous (storage)
13 / 51
Who moves? When do they move? All else equal, migration more likely: • Among agents with higher z • In the lean season • When experienced
Policy functions for migration take the form of thresholds in: • The transitory shock, s • Asset holdings, a
Model flexible; let experimental data disciplines whether e.g. agents migrate when assets sufficiently high or sufficiently low Value Functions
14 / 51
Calibration
15 / 51
Calibration Overview Pre-assign some parameters Set remaining parameters using simulated method of moments 1. Cross-sectional moments: e.g. urban-rural wage gap, rural share, variances of consumption and earnings. 2. Experimental moments: Perform the Bryan, Chowdhury, Mobarak (BCM) (2014) experiment in model.
16 / 51
Pre-Assigned Parameters
Parameter
Value
Source
Time period
Half year
—
Risk aversion, α
2.0
—
Discount factor, β
0.95
—
Gross real interest rate, R
0.96
1/ gross inflation rate
Rural seasonal productivity, Arl /Arg 50% drop in rural income Khandker (2012) Seasonal moving costs, mT
10% of rural consumption BCM (2014)
Permanent moving costs, mp
2 × mT
Very few permanent moves
Decreasing returns in rural area, φ
0.91
Akram, Chowdhury, Mobarak (2017)
17 / 51
Parameters to Calibrate Remaining parameters to calibrate. . . • Productivity urban area: Au • Shape parameter, urban permanent productivity: θ • Standard deviation of transitory shocks: σs • Urban relative risk parameter: γ • Persistence of transitory shocks: ρ • Disutility of migration: u¯ • Probability of gaining experience: 1 − λ • Probability of losing experience: 1 − µ
plus two sources of measurement error. . . to match 10 moments.
18 / 51
Calibration: The BCM (2014) Field Experiment The BCM (2014) field experiment • 100 randomly selected villages; 19 households in each village • Treatment group given 600 Taka ($8.50) conditional on migration; 200
Taka if they reported in at the destination. • Around two weeks of average rural wage
19 / 51
Calibration: The BCM (2014) Field Experiment The BCM (2014) field experiment • 100 randomly selected villages; 19 households in each village • Treatment group given 600 Taka ($8.50) conditional on migration; 200
Taka if they reported in at the destination. • Around two weeks of average rural wage
Key results/calibration targets: • 22 percent increase in seasonal migration in treatment relative to control
(58% vs 36%). • 11% increase in migration in the subsequent year (47% vs 36%). • Induced migrants increased their consumption by 30% relative to the
average household (i.e. the LATE of migration on consumption). 19 / 51
Performing the BCM (2014) Experiment in the Model • Randomly sample rural households in the stationary distribution • Draw from bottom half of rural asset distribution, as in experiment • Assign half to treatment, half to control • Solve for optimal policies, treating this offer as one-time, unanticipated,
with no general equilibrium effects. • Treatment group offered mT , if they move. • Compute take up of offer. Compute wages of those that take up.
20 / 51
Calibration Results — Targeted Moments
Moments
Data
Model
Control: Seasonal migrants
36
36
Control: Consumption increase of migrants (OLS)
10
10
Experiment: Seasonally migration relative to control
22
22
Experiment: Percent of treatment group that seasonally migrate in year 2
9
6
Experiment: Consumption of induced migrants relative to control (LATE)
30
30
1.80
1.80
63
63
Variance of log wages in urban
0.68
0.68
Variance of log consumption growth in rural
0.18
0.18
47
47
Urban-Rural wage gap Percent in rural
Percent of rural households with no liquid assets
21 / 51
Migration Rates: Data and Model
Migration Rate: Treatment Minus Control
30 Data 95-5 Confidence Interval (Data) Model Prediction
25
20
15
10
5
0
-5 2008
2009
2010
2011
2012
2013
22 / 51
Calibration Results — Parameters
Parameter
Value
Migration disutility, u¯
1.45
Probability gaining experience, 1 − λ
0.38
Probability losing experience, 1 − π
0.49
Shape parameter, urban talent, θ
2.08
Urban relative shock, γ
0.66
Productivity urban area Au
1.45
Standard deviation of transitory shocks
0.36
Persistence of transitory shocks
0.71
23 / 51
Migration Policy: Low z, Lean Season, No Experience
Assets
P.Migrate
S.Migrate
Stay
Zero Assets
0.16 0.2
0.27 0.35 0.45 0.59 0.77
1
1.3
1.7
2.21 2.88 3.75 4.89 6.37
Transitory Shock
24 / 51
Migration Policy: Moderate z, Lean Season, No Experience
Assets
P.Migrate
S.Migrate
Stay
Zero Assets
0.16 0.2
0.27 0.35 0.45 0.59 0.77
1
1.3
1.7
2.21 2.88 3.75 4.89 6.37
Transitory Shock
25 / 51
Migration Policy: Moderate z, Lean Season, Experience
Assets
P.Migrate
S.Migrate
Stay
Zero Assets
0.16 0.2
0.27 0.35 0.45 0.59 0.77
1
1.3
1.7
2.21 2.88 3.75 4.89 6.37
Transitory Shock
26 / 51
The Role of Credit Constraints • Workers “stuck” in rural areas due to credit constraints?
• Policy functions in previous slides suggested otherwise.
- Low income, low asset households are more likely to migrate. - Consistent with negative selection in OLS vs. LATE estimates.
• Additional test: unconditional cash transfer
- BCM ran unconditional transfer experiment too; found negligible effects on migration - Model: negligible effects on migration
27 / 51
The Role of Differential Risk • Urban area is risky and insurance is incomplete, so stay in rural?
(e.g Harris and Todaro, 1970)
• Suggests consumption of movers should be more variable than stayers
• But standard deviation of log consumption growth in fact similar:
Control Group Stay
Treatment Group
Migrate
Stay
Migrate
Data
0.39
0.43
Data
0.41
0.43
Model
0.41
0.43
Model
0.40
0.42
28 / 51
Calibration Results — Identification
Moments
Data Baseline θ = ∞ u¯ = 1 λ = 0
Migration, Control
36
36
48
29
37
Migration, Experiment
22
22
31
10
16
Migration, Experiment, year 2
9
6
0
0
0
Consumption, OLS
10
10
51
2
10
Consumption, LATE
30
30
20
6
30
1.80
1.80
1.80
1.80
1.80
63
63
57
89
92
Urban-Rural wage gap Percent in rural
29 / 51
Calibration Results — Identification
Moments
Data Baseline θ = ∞ u¯ = 1 λ = 0
Migration, Control
36
36
48
29
37
Migration, Experiment
22
22
31
10
16
Migration, Experiment, year 2
9
6
0
0
0
Consumption, OLS
10
10
51
2
10
Consumption, LATE
30
30
20
6
30
1.80
1.80
1.80
1.80
1.80
63
63
57
89
92
Urban-Rural wage gap Percent in rural
Migration rates help discipline productivity distribution A larger θ means less talent dispersion, more “marginal” households, more migration. Thus, migration reveals how many are on the margin or not. 29 / 51
Calibration Results — Identification
Moments
Data Baseline θ = ∞ u¯ = 1 λ = 0
Migration, Control
36
36
48
29
37
Migration, Experiment
22
22
31
10
16
Migration, Experiment, year 2
9
6
0
0
0
Consumption, OLS
10
10
51
2
10
Consumption, LATE
30
30
20
6
30
1.80
1.80
1.80
1.80
1.80
63
63
57
89
92
Urban-Rural wage gap Percent in rural Returns to Migration inform u. ¯
Consumption of new migrants must “jump” to compensate for the dislike of migration. 29 / 51
Calibration Results — Identification
Moments
Data Baseline θ = ∞ u¯ = 1 λ = 0
Migration, Control
36
36
48
29
37
Migration, Experiment
22
22
31
10
16
Migration, Experiment, year 2
9
6
0
0
0
Consumption, OLS
10
10
51
2
10
Consumption, LATE
30
30
20
6
30
1.80
1.80
1.80
1.80
1.80
63
63
57
89
92
Urban-Rural wage gap Percent in rural
Migration in year 2 informs dynamics of u. ¯ Some induced migrants keep migrating.
29 / 51
Calibration Results — Identification
Moments
Data Baseline θ = ∞ u¯ = 1 λ = 0
Migration, Control
36
36
48
29
37
Migration, Experiment
22
22
31
10
16
Migration, Experiment, year 2
9
6
0
0
0
Consumption, OLS
10
10
51
2
10
Consumption, LATE
30
30
20
6
30
1.80
1.80
1.80
1.80
1.80
63
63
57
89
92
Urban-Rural wage gap Percent in rural
Multiple ways to match urban-rural wage gaps. ⇒ Urban-rural wage gap by itself hard to interpret
29 / 51
Welfare
30 / 51
Consumption-Equivalent Welfare Gains From Experiment
2
Welfare Gains
1.5
1
0.5
0 0
10 25 40
100 50
83
65 66
80
Income Percentile
100
50, Zero Assets
Asset Percentile
31 / 51
Welfare From Unconditional Transfer (Same Cost as Experiment)
2
Welfare Gains
1.5
1
0.5
0 0
10 25 40
100 50
83
65 66
80
Income Percentile
100
50, Zero Assets
Asset Percentile
32 / 51
Increase in Migration: Treatment - Control
Why? The Poor Respond More to the Migration Incentive
60 50 40 30 20 10 0 0
10 25 40
100 50
83
65 66
80
Income Percentile
100
50, Zero Assets
Asset Percentile
33 / 51
Consumption-Equivalent Welfare Gains
Income Quintile
Conditional Migration Transfer
Unconditional Transfer
Welfare
Welfare | Migr.
Migr. Rate
Welfare
Migr. Rate
1
1.01
1.17
85.8
0.88
46.0
2
0.35
0.59
59.1
0.46
35.5
3
0.21
0.43
48.8
0.34
32.8
4
0.13
0.30
40.9
0.26
28.7
5
0.07
0.20
35.8
0.18
28.5
0.35
0.65
56.0
0.42
36.0
Average
34 / 51
Consumption-Equivalent Welfare Gains
Income Quintile
Conditional Migration Transfer
Unconditional Transfer
Welfare
Welfare | Migr.
Migr. Rate
Welfare
Migr. Rate
1
1.01
1.17
85.8
0.88
46.0
2
0.35
0.59
59.1
0.46
35.5
3
0.21
0.43
48.8
0.34
32.8
4
0.13
0.30
40.9
0.26
28.7
5
0.07
0.20
35.8
0.18
28.5
0.35
0.65
56.0
0.42
36.0
Average
35 / 51
Consumption-Equivalent Welfare Gains
Income Quintile
Conditional Migration Transfer
Unconditional Transfer
Welfare
Welfare | Migr.
Migr. Rate
Welfare
Migr. Rate
1
1.01
1.17
85.8
0.88
46.0
2
0.35
0.59
59.1
0.46
35.5
3
0.21
0.43
48.8
0.34
32.8
4
0.13
0.30
40.9
0.26
28.7
5
0.07
0.20
35.8
0.18
28.5
0.35
0.65
56.0
0.42
36.0
Average
36 / 51
Migration Disutility Important for Welfare Results
Income Quintile
Welfare Gains: Surprise No u¯ After Migration Welfare
Welfare | Migr.
Welfare | Induced
1
3.40
3.84
3.99
2
1.69
2.61
3.05
3
1.12
2.14
2.77
4
0.89
1.83
2.52
5
0.48
1.29
2.30
1.51
2.60
3.21
Average
37 / 51
Lack of High z Rural Workers Important for Welfare Results • Counterfactual: send representative set of urban households to rural area
– These have the relatively highest z draws in equilibrium – Also give them average assets of rural households – Re-simulate the experiment (still in partial equilibrium) • Welfare gains for this high-z group: 3.3% consumption equivalent
– Virtually all urban again within a few periods
38 / 51
Additional Evidence: Discrete Choice Experiments
39 / 51
Discrete Choice Migration Experiment
• New surveys of same villages as BCM (2014); conducted summer 2015
• Present two hypothetical migration options for 2015 lean season (fall
2015) to each respondent; pick Choice #1, Choice #2, or “No Migration”
• Options vary with respect to risk, amenities and wages at destination
• Goal: gauge importance of “migration disutility” relative to other
migration determinants
40 / 51
Discrete Choice Migration Experiment
S.1.C.2 Given the attributes below, which option do you choose? Please evaluate each new pair of migration options independent of the ones you saw earlier. Choice #1: Migration
Choice #2: Migration
Choice #3: No Migration
Chance of Employment
33%
33%
N/A
Daily Wage (Taka)
270
340
Wage at Home in November
Latrine Facility during Migration
Pucca Latrine in Residence
Walk to Open Defecate or Public Pay Toilet
N/A
Family Contact
See Family Every Month
See Family Every 2 Month
N/A
s16bq2_1 Your Choice (Tick Single Box)
Easy
Somewhat
Very 41 / 51
Estimated Marginal Effects on Migration Estimated Marginal E↵ects on Migration Table 1: Marginal Effects Of Migration Conditions on Migration Condition Migration Opportunity 1
Migration Opportunity 2
ME 0.000 (.)
PP 0.597*** (0.053)
66 Percent Chance of Employment
0.067*** (0.012)
–0.049*** (0.010)
0.732*** (0.046)
0.135*** (0.030)
0.200*** –0.086*** (0.045) (0.029)
100 Percent Chance of Employment
0.048*** (0.009)
–0.068*** (0.012)
0.791*** (0.040)
0.193*** (0.033)
0.161*** –0.125*** (0.038) (0.033)
Family visit once in 60 days
0.071*** (0.014)
0.000 (.)
0.760*** (0.041)
0.000 (.)
Family visit twice in 60 days
0.067*** (0.012)
–0.004 (0.008)
0.732*** (0.046)
–0.027 (0.024)
0.200*** 0.032 (0.045) (0.023)
Family visit 4 times in 60 days
0.058*** (0.012)
–0.013* (0.007)
0.763*** (0.049)
0.003 (0.028)
0.179*** 0.010 (0.046) (0.028)
Walk to Open Defecate or Public Pay Toilet
0.067*** (0.012)
0.000 (.)
0.732*** (0.046)
0.000 (.)
Pucca Latrine in Residence
0.029*** (0.006)
–0.038*** (0.008)
3449
–0.001*** (0.000) 3449
1
Raw Daily Wage(Taka) - Migration Opportunity 2 Observations
ME 0.000 (.)
At Home
PP 0.116*** (0.018)
33 Percent Chance of Employment
0.906*** (0.021)
0.174*** (0.032)
3449
0.004*** (0.000) 3449
PP 0.286*** (0.055)
0.169*** (0.040)
0.200*** (0.045)
ME 0.000 (.)
0.000 (.)
0.000 (.)
0.065*** –0.136*** (0.019) (0.032)
3449
–0.002*** (0.000) 3449
Standard errors in parentheses. * p < .10, ** p < .05, *** p < .01. Standard errors are adjusted for 2,566 clusters in hhid. PP columns represent predicted probabilities of migrating at given condition, and ME columns represent marginal effects of chaning migration conditions in each category. PP and ME are measured while fixing 1st migration conditions (wage, employment chance, family visit, latrine) at the worst, and fixing 2nd migration condition at median. Analysis sample includes only those households in the control group.
30 / 47
42 / 51
Estimated Marginal Effects on Migration Estimated Marginal E↵ects on Migration Table 1: Marginal Effects Of Migration Conditions on Migration Condition Migration Opportunity 1
Migration Opportunity 2
ME 0.000 (.)
PP 0.597*** (0.053)
66 Percent Chance of Employment
0.067*** (0.012)
–0.049*** (0.010)
0.732*** (0.046)
0.135*** (0.030)
0.200*** –0.086*** (0.045) (0.029)
100 Percent Chance of Employment
0.048*** (0.009)
–0.068*** (0.012)
0.791*** (0.040)
0.193*** (0.033)
0.161*** –0.125*** (0.038) (0.033)
Family visit once in 60 days
0.071*** (0.014)
0.000 (.)
0.760*** (0.041)
0.000 (.)
Family visit twice in 60 days
0.067*** (0.012)
–0.004 (0.008)
0.732*** (0.046)
–0.027 (0.024)
0.200*** 0.032 (0.045) (0.023)
Family visit 4 times in 60 days
0.058*** (0.012)
–0.013* (0.007)
0.763*** (0.049)
0.003 (0.028)
0.179*** 0.010 (0.046) (0.028)
Walk to Open Defecate or Public Pay Toilet
0.067*** (0.012)
0.000 (.)
0.732*** (0.046)
0.000 (.)
Pucca Latrine in Residence
0.029*** (0.006)
–0.038*** (0.008)
3449
–0.001*** (0.000) 3449
1
Raw Daily Wage(Taka) - Migration Opportunity 2 Observations
ME 0.000 (.)
At Home
PP 0.116*** (0.018)
33 Percent Chance of Employment
0.906*** (0.021)
0.174*** (0.032)
3449
0.004*** (0.000) 3449
PP 0.286*** (0.055)
0.169*** (0.040)
0.200*** (0.045)
ME 0.000 (.)
0.000 (.)
0.000 (.)
0.065*** –0.136*** (0.019) (0.032)
3449
–0.002*** (0.000) 3449
Standard errors in parentheses. * p < .10, ** p < .05, *** p < .01. Standard errors are adjusted for 2,566 clusters in hhid. PP columns represent predicted probabilities of migrating at given condition, and ME columns represent marginal effects of chaning migration conditions in each category. PP and ME are measured while fixing 1st migration conditions (wage, employment chance, family visit, latrine) at the worst, and fixing 2nd migration condition at median. Analysis sample includes only those households in the control group.
30 / 47
43 / 51
Estimated Marginal Effects on Migration Estimated Marginal E↵ects on Migration Table 1: Marginal Effects Of Migration Conditions on Migration Condition Migration Opportunity 1
Migration Opportunity 2
ME 0.000 (.)
PP 0.597*** (0.053)
66 Percent Chance of Employment
0.067*** (0.012)
–0.049*** (0.010)
0.732*** (0.046)
0.135*** (0.030)
0.200*** –0.086*** (0.045) (0.029)
100 Percent Chance of Employment
0.048*** (0.009)
–0.068*** (0.012)
0.791*** (0.040)
0.193*** (0.033)
0.161*** –0.125*** (0.038) (0.033)
Family visit once in 60 days
0.071*** (0.014)
0.000 (.)
0.760*** (0.041)
0.000 (.)
Family visit twice in 60 days
0.067*** (0.012)
–0.004 (0.008)
0.732*** (0.046)
–0.027 (0.024)
0.200*** 0.032 (0.045) (0.023)
Family visit 4 times in 60 days
0.058*** (0.012)
–0.013* (0.007)
0.763*** (0.049)
0.003 (0.028)
0.179*** 0.010 (0.046) (0.028)
Walk to Open Defecate or Public Pay Toilet
0.067*** (0.012)
0.000 (.)
0.732*** (0.046)
0.000 (.)
Pucca Latrine in Residence
0.029*** (0.006)
–0.038*** (0.008)
3449
–0.001*** (0.000) 3449
1
Raw Daily Wage(Taka) - Migration Opportunity 2 Observations
ME 0.000 (.)
At Home
PP 0.116*** (0.018)
33 Percent Chance of Employment
0.906*** (0.021)
0.174*** (0.032)
3449
0.004*** (0.000) 3449
PP 0.286*** (0.055)
0.169*** (0.040)
0.200*** (0.045)
ME 0.000 (.)
0.000 (.)
0.000 (.)
0.065*** –0.136*** (0.019) (0.032)
3449
–0.002*** (0.000) 3449
Standard errors in parentheses. * p < .10, ** p < .05, *** p < .01. Standard errors are adjusted for 2,566 clusters in hhid. PP columns represent predicted probabilities of migrating at given condition, and ME columns represent marginal effects of chaning migration conditions in each category. PP and ME are measured while fixing 1st migration conditions (wage, employment chance, family visit, latrine) at the worst, and fixing 2nd migration condition at median. Analysis sample includes only those households in the control group.
30 / 47
44 / 51
Conclusions and Future Work
• General-equilibrium model of migration; novelty: disciplined using
experimental evidence
• Welfare gains from migration subsidies mostly from targeting poorest
• Not about reducing misallocation of workers across space
• Future work: comparison to rural employment guarantees; effects of
scaling up migration subsidies
45 / 51
Extra Slides
46 / 51
Rural Household’s Problem back
Problem of household in rural area with productivity z: v (a, r , s, x, i, Nr ) = max v (a, r , s, x, i, Nr | stay), v (a, r , s, x, i, Nr | perm), v (a, r , s, x, i, Nr | seas)
47 / 51
Rural Household’s Problem back
Problem of household in rural area with productivity z: v (a, r , s, x, i, Nr ) = max v (a, r , s, x, i, Nr | stay), v (a, r , s, x, i, Nr | perm), v (a, r , s, x, i, Nr | seas)
Valuing of staying in rural area, v (a, r , s, x, i, Nr | stay) is:
max u(Ra + wr (s, i, Nr ) − a0 ) + β E[v (a0 , r , s 0 , x 0 , i 0 , Nr0 )]
a0 ∈A
47 / 51
Rural Household’s Problem back
Problem of household in rural area with productivity z: v (a, r , s, x, i, Nr ) = max v (a, r , s, x, i, Nr | stay), v (a, r , s, x, i, Nr | perm), v (a, r , s, x, i, Nr | seas)
The value of permanently moving, v (a, r , s, x, i, Nr | perm), is: 0 0 0 0 0 0 max u(Ra + w r (z, s, i, Nr ) − a − mp ) + β E[v (a , u, s , x , i , Nr )] 0
a ∈A
47 / 51
Rural Household’s Problem back
Problem of household in rural area with productivity z: v (a, r , s, x, i, Nr ) = max v (a, r , s, x, i, Nr | stay), v (a, r , s, x, i, Nr | perm), v (a, r , s, x, i, Nr | seas)
The value of seasonally moving, v (a, r , s, x, i, Nr | seas), is:
max u(Ra + wr (s, i, Nr ) − a0 − mT ) + β E[v (a0 , seas, s 0 , x 0 , i 0 , Nr0 )]
a0 ∈A
where v (a0 , seas, s 0 , x 0 , i 0 , Nr0 ) is: 0 00 x 00 00 00 00 00 max u(Ra + w ¯ + β E[v (a , r , s , x , i , N )] u (z, s) − a )u r 00 a ∈A
47 / 51
The Role of Information • BCM did an information experiment too
- Treatment group instruction on types of jobs in urban areas - Also information on average wages, and where/how to find these jobs - Result: precise zero effect on migration
• We did follow-up surveys on same villagers on wage expectations, 2014
- Ratio of perceived Dhaka wages to rural Rangpur wages = 2.4 - Averages from Household Income and Expenditure Survey = 2.2
• Consistent with model’s assumption of rational expectations
48 / 51
Wage Expectations
Section 15: Wage Expectations Enumerators: Please tell the respondent to answer these questions for employment inclusive of any fringe benefits. For example, if an agricultural laborer cash-in-hand payment is 100 taka/day and also receives food valued at 20 taka/day their daily wage would be 120 taka/day.
Q# Question s15q1 How much do you think you would make per day if you were to stay at home in November?
Units Taka/Day
s15q2 How much do you think the average permanent resident of Dhaka makes per day in November?
Taka/Day
s15q3 How much do you think the average seasonal migrant in Dhaka makes per day in November?
Taka/Day
Response
s15q4 If you were to migrate to Dhaka this November, Taka/Day how much do you think you would make per day?
49 / 51
Fraction
.2
.3
Perceived Ratio of Dhaka Wages to Own Wages (2014)
0
.1
Mean = 2.42 Median = 2.33
0
1
2
3
4 Ratio
5
6
7
8
50 / 51
Actual Urban-Rural Wage Gaps in HIES (2010)
Mean
Median
All urban / All rural
1.80
1.72
All urban / Rangpur rural
2.12
1.87
Dhaka / Rangpur rural
2.20
2.01
51 / 51